Cooling The Machines, Heating The Planet: The Hidden Water Cost Of Large Language Models (LLMs) – OpEd

When you ask LLMs such as ChatGPT to draft an email or Gemini to summarize an article, or DeepSeek to review a paper, you are tapping into the power of Large Language Models (LLMs). The LLMs and AI systems are trained on massive amounts of text data to mimic human language. These models, like OpenAI’s GPT-4, Google’s Gemini, DeepSeek-R1 and Meta’s Llama 2, are revolutionizing industries, from healthcare to education. But behind their sleek interfaces lies a secret their insatiable thirst for water. Would you use ChatGPT, Gemini, Llama, DeepSeek and other LLMs if you knew each response cost a glass of water?

The training and operation of LLMs require massive computational power, which generates lots of heat. To prevent this overheating, the data centers rely on cooling systems that consume staggering amounts of water. For instance, training OpenAI’s GPT-3 required approximately 700,000 liters of water enough to fill a nuclear reactor’s cooling tower. Google’s data centers, which power models like Gemini, consumed 5.6 billion gallons of water in 2022, largely for cooling. That’s equivalent to the daily water needs of 43,000 households. Even Meta’s Llama 2 training consumed 10.9 millions liters

The problem doesn’t end with training. Every time you interact with an LLM, it consumes water. For example, every 10-50 prompts to ChatGPT use roughly 500 milliliters of water, depending on server load. Google Search, which processes 8.5 billion queries daily, uses 10 times more water when enhanced by AI. The location also plays a critical role: data centres in hot, dry regions like Arizona or Chile use 40% more water for cooling than those in cooler climates. Microsoft’s Arizona data center, for instance, uses 56 million liters of water annually enough to sustain 1,400 families for a year.

The environmental and social consequences of this water usage are profound. In drought-prone regions, data centers compete with agriculture and drinking water for limited resources. In Chile, for example, data centers have been accused of exacerbating water shortages for local farmers. Meanwhile, the energy required to power these systems often comes from coal or gas plants, which themselves consume vast amounts of water for cooling. ChatGPT’s monthly energy use indirectly consumes million liters of water through power generation alone.

The social justice implications are equally alarming. Data centers are frequently built in low-income or marginalized areas, creating “sacrifice zones” where communities lack the resources to protest. In Oregon’s “Data Center Alley,” residents have raised concerns about the environmental and health impacts of nearby facilities. Globally, the water footprint of AI exacerbates inequality, with drought-stricken regions in the Global South hosting servers for Western tech firms. Kenya, for example, faces severe water shortages, yet its resources are being diverted to cool machines that serve distant economies.

Despite these challenges, tech companies often tout their commitment to sustainability. Google, for instance, claims its AI is “carbon-neutral,” but says little about its water usage. This hypocrisy underscores the need for greater accountability. The water crisis won’t be solved by algorithms it demands human action. Tech giants must prioritize sustainable practices, such as relocating data centers to cooler climates, investing in air-cooled systems, and recycling wastewater. Governments must regulate AI’s environmental footprint as rigorously as its ethics. And as users, we must ask ourselves: Do we need another AI-generated poem if it means a child goes thirsty?

LLMs are reshaping our world, but their hidden water costs threaten to drain our future. We cannot let Silicon Valley’s race for smarter AI outpace our planet’s ability to survive it. The water crisis is a human crisis, and it demands human solutions. As we marvel at AI’s capabilities, let’s not forget the real cost of its intelligence: the water we drink, the air we breathe, and the future we share.